[time-nuts] Clocks for Audio gear

Hal Murray hmurray at megapathdsl.net
Wed May 9 21:25:34 UTC 2012


was Subject: Re: [time-nuts] Faster than light of a different type
(Probably my fault.)

actast at hotmail.com said:
> What I found funny was that the Audiophlie and light thread drew such
> attacks when it hit home to me as exactly what the Time-Nuts mission is
> about.  The Audio thread touched on some real world time and freq research
> ...

I too enjoyed the technical discussions.  Thanks for your contributions.

It's the audiophool bashing that people are complaining about.  Sure, it's 
fun, but only at the right time and it gets old quickly.  The problem is that 
with large groups, there are different opinions of when and how much is 
appropriate.  The long tail on opinions of reasonable can annoy a lot of 
people.

-----------

Back to technical stuff...

As a practical matter, is clock jitter or phase noise from a typical low cost 
crystal and decent board layout a significant problem in audio gear?  How 
hard is it to measure?

Is clock accuracy a practical problem?  How good are people with perfect 
pitch?  It wouldn't surprise me if there are a few that are much much better 
than others, but how good is that relative to 50 PPM which I can get in a low 
cost crystal?

Video geeks solved their clock distribution problems by using frame buffers.  
Is there a similar trick for audio?  Is there a need for it?

--------

I know clocking is a serious problem in fancy DSP systems.  For example, 
modern radar has gone digital.  In that context, clock jitter can be 
important.  Standard procedure is don't run your clock through a FPGA because 
it will add jitter.

Part of the problem is that they are doing magic down conversion in the ADC.  
(I can't think of the term.)  Suppose you have a 100 MHz signal with a 1 MHz 
bandwidth.  You don't have to sample at 200 MHz.  You can sample at 2 MHz and 
your signal will alias down.  It's turning what is normally a bug into a 
feature.  The catch is that the errors/noise due to clock jitter happens at 
the high frequency, in this case multiplying the noise by 100.  (Your 
sample/hold at the front end has to work at the high frequency and your 
anti-aliasing filter gets more interesting.)

There has been an interesting change in the specs for ADCs and DACs over the 
past 20(?) years.  They used to be specified using terms like DNL and INL and 
No-missing-codes.  Modern high-speed ADCs are specified with terms like ENOB 
and SFDR.  Data sheets often include several plots of a batch of samples run 
through a DFT so you can see the noise floor and such.

Here is a reasonable glossary:
  http://pdfserv.maxim-ic.com/en/an/AN641.pdf

I don't remember comments/specs about clock jitter in the data sheets but I 
haven't looked at one in a few years.  I'll have to keep an eye out the next 
time I'm browsing.



-- 
These are my opinions.  I hate spam.






More information about the time-nuts mailing list